When we started making tests of the L2A production at MUSCATE with Sentinel-2B data, we considered using a small correction of Sentinel-2B to correct for a eventual bias between the reflectances of each satellite. Such differences were observed by colleagues form CNES, and were comprised between 0 and 2.5%. Finally, we decided not to implement this in the operational line, because the figures we had were based on a small duration of acquisitions and were possibly not accurate enough.
Due to a human error, it turns out that the parameter file that we used for these tests made its way to the operational line. MAJA parameters are kept under a GIT version manager, which should reduce the probablility of such errors, but that parameter file is a parameter of MUSCATE, not MAJA, which was not managed yet. And as you know, according to Mr Murphy, when something can turn wrong, it turns wrong. the test calibration adjustment made its way to the production line.
For the Sentinel-2B L2A products produced until the 30th of November, the L1C TOA reflectance values were divided by the coefficients provided in the table above. On the first of December, this error was corrected. As the differences are quite small, we have not removed the Sentinel-2B products produced before December 1st, but we are going to reproduce them and replace them by the correct values during the coming weeks.
We are sorry for hat error, which will make us improve our verification procedure.